GitLab

This is the equivalent to the distribution_image for users of the
FDO_REPO_SUFFIX. Given a FDO_REPO_SUFFIX used for the container build stage,
this template provides the required variables to use that image (or the
path to it) in other jobs.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

Because it is useful to have around.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

We have container-build which pushes the image "somewhere" but those that will
use that image need to re-build that same path. Let's provide a template to
use a specific image instead.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

`/cache` is often used to speed up compilation with `ccache`. We should
allow the users of CI templates to mount this directory if it exists
so they can set up their runners as they wish and benefit from `ccache`.

Note that to allow the CI to test that we can mount the `/cache` folder.
So we introduce a new variable `CACHE_DIR`, which defaults to `/cache`,
that allows to tell which folder needs to be mounted as `/cache` in the
`buildah run $DISTRIBUTION_EXEC` stage.
- this allows to overwrite the `/cache` folder if it already exists in
  the current runner
- this also allows to export its content as an artifact in the CI job

Fixes #5Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

The `before_script` in the main container build was not that useful:
- `*fdo_distro_vars` was set but not used (and rewritten in `script`)
- the login was there, which allowed us to not have it in the `script`

However, this means anybody who wants to rewrite a `before_script` needs
to remember to also log in to the registry. Better ensure we do it in the
script so that we do not add burden on users.
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

Fixes #10Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

New convention is: .fdo.<job>@<distribution>[@<arch>]
which puts the thing to do at the most significant place and then the ability
to rotate through distributions at the LSBs.

This patch only affects the template itself, the CI scripts we use in this
repo are left with the same previous naming system - mostly because of
laziness.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

Where the templates get used it becomes tricky to figure out which extends:
section refers to a ci-templates one and which one is a local one. So let's
use a consistent prefix for all jobs and variables.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

This is an API break, previous users of FEDORA_TAG or similar must update.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

This is an API break, previous users of DEBIAN_EXEC or similar must update to
the new variable.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

This is an API break, previous uses of e.g. FEDORA_VERSION will stop
working and all users are required to switch to the new DISTRIBUTION_VERSION
instead.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

Downstream users of the ci-templates don't need to care whether packages are
rpms, debs or something else. And they shouldn't need separate variables per
distribution to install the same package set.

This is an API break, previous users of FEDORA_RPMS and friends will stop
working.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

This just the most basic scaffolding and build script to hook up to gitlab
pages. Actual docs need to be added.

The build script is a minimal wrapper around the sphinx-build command but this
way we can easily ensure that local rebuilding of the docs produces the same
output that gitlab pages will display.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

The point of this is to install a package for testing purposes and some
distros have wget and others have curl as part of the base image. So rather
than making the templates complicated, just install both and call both - one
of the two will prove the point.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

The upstream docker image uses 'archlinux' but we use just 'arch' everywhere
else in the templates. This means we need to have special handling for arch to
get the CI repository path correctly where we're processing multiple
distributions the same way.

Let's not do that, just drop the use of archlinux and use just "arch" even for
our repository path. I somewhat doubt anyone cares about matching the
docker hub path anyway.

This will rebuild the arch image for everyone who updates the templates.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

And where the image unpack fails too, unlikely though that is.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

There is a chance we try to compress the image before the VM has stopped,
which means we won't be able to start the base mkosi image next time
as it will fail while checking the file-system.
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

And bump also the normal container base image so we have something newer
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

There is no point in pulling the weak deps most of the time, so try to
stick with smaller images
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

As of today, the fedora:rawhide docker image still points
to f32, when it should be f33.

Manually force the versions.
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

Ubuntu 18.10 is now EOL, and they pulled in the plug on the servers,
so the jobs are failing.
Update to 19.10
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

This ensures that the release and GPG keys used match the rest of the
packages we're about to upgrade to.
Signed-off-by: Bastien Nocera <hadess@hadess.net>

It was missing an "s"
Signed-off-by: Bastien Nocera <hadess@hadess.net>

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

This needs to be defined as global variable in the actual project so we can
fetch our images from there. Where it isn't defined, the skopeo command will
fail. Let's at least warn about it.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

The exact message can be adjusted in src/defaults.yml.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

During execution of distro commands (e.g. $DEBIAN_EXEC), make sure that
we preserve the value of $HOME for execution in the container.

By default, newer versions of buildah seem to strip the environment
completely. This results in $HOME being empty on entry to the container.

If a build container is later reused for native build steps through the
GitLab Docker executor, $HOME will be preserved. When using things which
depend on the value of $HOME, like Python's pip, we can end up with
different expectations. In particular, pip will install into
$HOME/.local/bin, but since $HOME is $HOME, we can't later find what it
actually installed.
Signed-off-by: Daniel Stone <daniels@collabora.com>

The `check commit`  job is not present on master, and this leads to
https://gitlab.freedesktop.org/wayland/ci-templates/pipelines/87523

-> big failure. Remove the dependency on `check commit` and revisit later
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

alpine is not supported by mkosi
centos is having troubles too?
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

This image will be used to run mkosi which will generate the various
final VM images. However, the final container image will not be based
on this container, but on the qemu one.
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

It turns out that it is possible to start a VM directly from
an unprivileged container. So provide a new base image that we can
then use to create and launch new qemu images to be able to run
test suites that could not be run in a plain container because they
needed access to the kernel.
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

The `check commit`  job is not present on master, and this leads to
 https://gitlab.freedesktop.org/wayland/ci-templates/pipelines/85923

-> big failure. Remove the dependency on `check commit` and revisit later
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

We should probably save some runner resources if the very initial stages
are failing.

It´ s a shame gitlab´s DAG implementation forces us to manually keep track
of all requirements, but heh.
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

Previously, we were always using `buildah:latest`, which introduced the
side effect that over time, projects will not be using the bootstrap
we have been conducting CI.

Thanks to the templating, it´ s now easy enough to define the tags in
defaults.yml and reuse it everywhere.
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>